The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous examples for executing powerful Azure CLI commands that modify the state of an Azure DevOps organization.
Evidence: Commands in references/org-and-security.md allow for managing users (az devops user add/remove), security groups (az devops security group create/delete), and security permissions (az devops security permission update/reset).
Evidence: The skill documents the installation of Azure DevOps organization extensions using az devops extension install in references/org-and-security.md.
[EXTERNAL_DOWNLOADS]: Fetches installation scripts from a remote source.
Evidence: SKILL.md includes a command to download and execute the Azure CLI installation script from Microsoft's official infrastructure: curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash.
[CREDENTIALS_UNSAFE]: The skill provides patterns for handling sensitive authentication tokens and service principal keys that involve local storage.
Evidence: references/workflows-and-patterns.md contains a script that writes service principal keys to a local JSON file (service-connection.json) before using it to create a service endpoint.
Evidence: SKILL.md and references/workflows-and-patterns.md describe using Personal Access Tokens (PAT) for authentication, including piping them to the az devops login command.
[DATA_EXFILTRATION]: Automation patterns include sending data to external endpoints.
Evidence: The monitor_pipeline function in references/workflows-and-patterns.md uses curl to send pipeline failure alerts and run IDs to an external Slack webhook.
[PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection where untrusted data is processed without sanitization.
Ingestion points: Untrusted data enters the context through git commit messages (git log -1 --pretty=%B), build environment variables (BUILD_BUILDNUMBER), and pipeline run metadata fetched from the API as seen in references/workflows-and-patterns.md.
Boundary markers: Automation scripts lack delimiters or sanitization logic when interpolating these variables into CLI command arguments.
Capability inventory: Processed data is used in commands that create pull requests (az repos pr create), work items (az boards work-item create), or send network requests (curl).
Sanitization: No escaping or validation of external content is present in the provided shell script examples.

azure-devops-cli